Herein, we propose a two-stage method for the global minimization of the conformational energy function of biomolecules. Various methods have been developed to determine the global minimum of the energy function from a large number of local minima. However, it is not very easy to achieve this using a single method because of the vast conformational space of a molecule. It seems effective to combine two methods. The first one is to locate a fairly narrow region that contains the global minimum, and the second is to determine the global minimum. At the first stage of our two-stage method described herein, the diffusion equation method (DEM) is used to locate the region that contains the global minimum. At the second stage, the energy surface along the eigenvectors of the Hessian around the solution of the DEM is calculated and the local minima with lower energies than the DEM solution are more efficiently searched than a random search. In some cases, this local minimum search is not enough. However, we successfully obtain a minimum, which seems to be the global minimum, by searching the local minima at each step of the time-reversing procedure of the DEM. This method is applied to the small DNA fragment, d(GCGAAGC), which is known to form a stable minihairpin structure, and to a model structure of an oligopeptide that consists of nine alanines. For the DNA fragment, the same structure that was determined using nuclear magnetic resonance spectroscopy (NMR), which cannot be found using the DEM alone, was obtained with this method. For the oligopeptide, the structure obtained using the DEM is a seven-gamma-turn structure, and an alpha helix structure with a lower conformational energy than that of the gamma turn structure is obtained using the two-stage method. This method will be applicable to larger molecules than those calculated here.